• Aucun résultat trouvé

Time-changes of self-similar Markov processes

N/A
N/A
Protected

Academic year: 2022

Partager "Time-changes of self-similar Markov processes"

Copied!
8
0
0

Texte intégral

(1)

A NNALES DE L ’I. H. P., SECTION B

J. V UOLLE -A PIALA

Time-changes of self-similar Markov processes

Annales de l’I. H. P., section B, tome 25, n

o

4 (1989), p. 581-587

<http://www.numdam.org/item?id=AIHPB_1989__25_4_581_0>

© Gauthier-Villars, 1989, tous droits réservés.

L’accès aux archives de la revue « Annales de l’I. H. P., section B » (http://www.elsevier.com/locate/anihpb) implique l’accord avec les condi- tions générales d’utilisation (http://www.numdam.org/conditions). Toute uti- lisation commerciale ou impression systématique est constitutive d’une infraction pénale. Toute copie ou impression de ce fichier doit conte- nir la présente mention de copyright.

Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques

http://www.numdam.org/

(2)

Time-changes of self-similar Markov processes

J. VUOLLE-APIALA

Aarhus University, Department of Mathematics, Ny Munkegade, DK-8000 Aarhus C., Denmark Ann. Inst. Henri Poincaré,

Vol. 25, n° 4, 1989, p. 581-587. Probabilités et Statistiques

ABSTRACT. - Let

Xt

be a

03B2-self-similar, 03B2

>

0,

transient Markov process

on

(0, oo).

We show that if

XTt (Tt

is the

right

continuous inverse of a

continuous additive functional

At)

is an 03B1-self-similar Markov process,

a > 0,

then

A result

concerning time-changes

of a transient

Levy

process is also

given.

Key words : Self-similar, Markov process, time-change, Levy process.

RESUME. - Soit

Xt

un processus

p-self-similaire

transient et de Markov

sur

(0, oo), ~3 > o.

Notons

Tt

l’inverse continu a droite du fonctionnelle additive

At.

Nous montrons que si

XTt

est un processus a-self-similaire et de

Markov, Cl>O,

alors

Un resultat concernant le

changement

de temps d’un processus de

Levy

transient est

également

donne.

(3)

582 J. VUOLLE-APIALA

0. INTRODUCTION

a-self-similar Markov processes

(a-ssmp)

on

(0, oo)

were introduced

by

J.

Lamperti [5].

The process

(Xt, P")

with a state space

(0, oo)

is called

a-ssmp,

ex>O,

if there exists a Borel

semigroup (Pi, ))t _> o

on

(0, oo ) (0, oo ) satisfying

(i) Po ( ~ ) = I~

(ii) Pt (x, aa A)

for all

t > 0, a > 0, x E (0, oo ), A E ~ (0, oo ),

such that

(Xt, Px)

is a time

homogeneous

Markov process with a transition

function

(Pt ( ,

and with

sample paths

that are P~-almost

surely right-continuous

with left limits for all

x E (0, oo).

It was

proved

in

[6]

that every a-ssmp with "nice

paths" (see Notation)

on

(0, oo )

has a weak dual with

respect

to the measure dx. In this note we

apply

this result and

characterize,

the

theory developed

in

[3]

as

a main

tool,

all the

possible

ways to

time-change

a transient self-similar process

(in fact,

an a-ssmp is transient iff it is

cotransient;

see

Proposition)

to another self-similar process. We also obtain a result

concerning

time-

changes

of a

Levy

process. For

simplicity,

we assume

a > 0,

but the results

can

easily

be

generalized

to

negative

a.

1. - NOTATION. DEFINITIONS

Q notes the space of all functions o from

[0,

~

(0, oo) U ~ ~ ~ (A

denotes the

point

used as a

"graveyard";

we assume A is an isolated

point),

which

satisfy

(a)

for

t>_0; ~(t)=0 ~;

(b) co

is

right

continuous and co or has left limits on

[0, oo)

at every Such Q is called the space of "nice

paths".

DEFINITION. - Let a > 0 be

given.

A stochastic process

(Xt,

with a

state space

(0,

is called a-ssmp on

(0, oo )

if the

following

is

satisfied: there exists a Borel

semigroup (Pt ( , ))t >_ o

on

(0, oo)

X ~

(0, oo)

with the

properties:

(i) 1’0 ( ~ ) = I~

(ii) Pt (x,

for all

t ~ 0, a > 0, x E (0, oo ), A E ~ (0, oo ),

such that

(Xt, P~)

is a time

homogeneous

Markov process with a transition function

(Pt ( , )t > o

and for

x E (0, ao ).

Remark I. -

It was

proved

in

[4]

that every a-ssmp o.n

(0, oo ) automati- cally

is

strongly

Markov.

A Markov process

(Xt, Px, x E (0, oo))

is said to be in weak

duality

with

a Markov process

(Xt, Px, x E (0, oo))

with

respect

to a a-finite measure

Annales de l’lnstitut Henri Poincaré - Probabilités et Statistiques

(4)

583

SELF-SIMILAR MARKOV PROCESSES

r~, if for all

bounded f, g

in

~ (o, oo)

Let

(Xt, P~)

be in weak

duality

with

(Xt, Px)

with

respect

to a measure

(Xt, PX)

is said to be

transient,

if

for all x, all

bounded, non-negative

Borel functions

f

on

(0, oo)

with

compact support (see

alternative definitions for transience in

[2]).

If the

dual process

(Xt, Px)

is

transient,

then

P~)

is said to be cotransient.

Remark 2. - It was shown in

[6]

that an a-ssmp on

(0, oo )

has a weak

dual,

with

respect

to the measure

xl~«- ~ dx,

and the dual process is also

an a-ssmp.

2. - THEOREMS

We assume

throughout

this paper that

P~)

is transient

(as

we shall

see in

Proposition,

for self-similar processes the transience is

equivalent

to the

cotransience). According

to

[6], (XTt, P~)

is an a-ssmp if

(Xt, P")

is

p-ssmp

and

Tt

is the

right

continuous inverse of an additive functional k We shall show that this is the

only possible

way to time-

change (Xt, Px)

to an a-ssmp.

PROPOSITION. - Let

(Xt, Px)

be a

03B2-ssmp

on

(0, r), [3 >

0. Then it is

transient

iff

it is cotrasient.

Proof. - According

to

[4] (Th. 2.3)

and

[5] (Th. 4.1),

there is one to

one

correspondence

between a

p-ssmp Xt

on

(0, oo)

and a

Levy

process

Zt

on

( - oo, + oo) (that is, Zt

is a

strong

Markov process which have

stationary independent

increments and

right

continuous

paths

with left

limits)

defined

by

where

Tr

is the

right

continuous inverse of

an additive functional

It is also

easily

seen that

Xt

is transient iff

Zr

is transient. Now for

Zt

there exists a weak dual

2~

with

respect

to the

Lebesgue

measure such

that also

Zt

is a

Levy

process. As shown in

[6], starting

from

Zt

one can

construct a

03B2-ssmp t,

which is a weak dual to

Xt

with respect to the

(5)

584 J. VUOLLE-APIALA

measure dx. Now

Xt

is transient iff

Zt

is transient and so it suffices

to show that

Zt

is transient iff it is cotransient. If

(Zt, QZ)

is a

Levy

process then

Zt

under

QZ

has the same distribution as

z + Zt

under

and thus easy calculations show that

Zt

under

QZ

has the same distribution

as

z - Zt

under

Q°.

This shows that

Zt

is transient iff

Zt

is transient and

gives

thus the assertion.

In the

proof

of the

following

theorem

actually

cotransience

(and

not

transience)

is used.

THEOREM 2.1. - Let

(Xt, Px)

be a transient

03B2-ssmp

on

(0, oo), (3

>

0,

and

let

At be a

continuous

additive functional

with

Tt

as the

right

continuous

inverse,

i. e.

If

the process

Px)

is 03B1-ssmp, then there exists k > 0 such that

Proof. -

As mentioned in Remark

2, (Xt, P")

has a weak dual

(Xt, P")

with

respect

to the measure

x l ~~ -1 dx

such that also

(Xt, P")

is

p-ssmp.

Let

At

be a continuous additive functional of

Xt

and let

(XTt, P"), Tt

is

the

right

continuous inverse of

At,

be an a-ssmp. Let further vA be the Revuz measure

corresponding

to

At. According

to the result of Getoor and

Sharpe [3]

for any

bounded, non-negative

Borel function

/

with

compact support.

The

right

side of

(2.1)

is

equal

to

~0f(XTt)dt }y1/03B2-1 dy, which,

because of the

a-self-similarity

of

Px),

is

equal

to

Annales de l’lnstitut Henri Poincaré - Probabilités et Statistiques

(6)

SELF-SIMILAR MARKOV PROCESSES 585

From

(2.1)

and

(2 . 2)

we

obtain, by substituting z = a°‘ y,

The

03B2-self-similarity

of

(Xt, Px) implies

This, together

with

(2. 3), gives

Because

Xt

is cotransient we have

LJ, f ’ (x)

+ oo , V x. Thus

Applying

the well-known

uniqueness

result for a Haar measure we obtain

which

gives

Remark. - The

special

case of Theorem 1 is oc =

P,

which

gives At

= kt.

This means that the

only possible

way to

time-change

a transient

P-ssmp

to another

P-ssmp

is a linear

time-change.

In

[5]

J.

Lamperti

introduced a continuous additive functional

where

(Xt, P~)

is a

p-ssmp

on

(0, oo).

He showed that if

Tt

is the

right

continuous inverse of

At,

then the

time-changed

process

(XTt, P~)

is a

strong Markov process on

(0, oo)

such that it is

multiplicatively invariant,

I. e.

A) = Qt (ax, a A),

for all

Xe(0, oo ), oo ),

where

Qt ( , )

is a transition function for

(XTt, Px).

The

following

theorem says that

this, possibly multiplied by

a constant, is the

only

way to

time-change (X~, P")

to a

multiplicatively

invariant

(7)

586 J. VUOLLE-APIALA

process:

THEOREM 2.2. - Let

Px)

be a transient

03B2-ssmp

on

(0, oo)

and let

At

be a continuous additive

functional of Px)

and

Tt

the

right

continuous

inverse

of Ar. If Px)

is

multiplicatively invariant,

then there exists k > 0 such that

The

proof

is similar to that of Theorem 2.1 and will therefore be omitted.

Finally,

we shall

present

a result

concerning time-changes

of a

Levy

process. It was

already

remarked

by Lamperti [5]

that

(Yt)

is a

multiplica- tively

invariant

strong

Markov process with "nice

paths" (see Notation)

on

(0, oo)

iff

(Zt) _ (log Yt)

is a

Levy

process on

( -

oo, -

oo).

We will

show

THEOREM 2.3. - The

only possible

way to

time-change

a transient

Levy

process on

( -

~, +

oo )

to another

Levy

process is a linear

time-change t - kt,

k>o.

We need the

following

Lemma:

LEMMA. - Let

(Yr, P~)

be a

transient, multiplicatively

invariant strong Markov process with "nice

paths" (see Notation)

on

(0, oo).

Then the

only possible

way to

time-change (Yt, Px)

to another process

of

the same type is

a linear

time-change

t -

kt,

k > o.

Proof. -

What we have to show

is,

that if

At

is a continuous additive functional of

(Yt)

with

Tt

as the

right

continuous inverse and

(YTt, P~)

is

multiplicatively invariant,

then there exists k > 0 such that

At = kt

for all

t ~. According

to

[6], (Yt, P")

has a weak dual with

respect

to the

measure dx. We can now

show, by

the same way as in Theorem

2.1,

that

At

has a Revuz measure

which

gives

the assertion.

Proof of

Theorem 2.3. - Let

(Zt, QZ)

be a transient

Levy

process on

( - oo, +00). Then,

as remarked in the

proof

of

Proposition,

the weak

dual

(Zt, QZ),

which also is a

Levy

process, is transient. Now

(exp Zt, "),

x >

0,

is

multiplicatively

invariant and

has,

as shown in

[6],

a weak dual

(expZt,

with

respect

to the measure dx. It is

easily

seen that

(Zt)

is transient iff

(exp Zt)

is transient and so,

according

to

Lemma, (exp Zt)

cannot be

time-changed

to another

multiplicatively

invari-

ant process with "nice

paths"

otherwise than

by

the linear time

change t - kt,

k > o. Thus one to one

correspondence

between the class of

Levy

Annales de l’lnstitut Henri Poincaré - Probabilités et Statistiques

(8)

SELF-SIMILAR MARKOV PROCESSES 587

processes and the class of

multiplicatively

invariant processes with "nice

paths" gives

the assertion.

REFERENCES

[1] R. M. BLUMENTHAL and R. K. GETOOR, Markov Processes and Potential Theory, New York, Academic Press, 1968.

[2] R. K. GETOOR, Transience and Recurrence of Markov Processes, Séminaire de Probabili- tés XIV, 1978/79, pp. 397-409.

[3] R. K. GETOOR and M. J. SHARPE, Naturality, Standardness and weak Duality for

Markov Processes, Z. Wahrscheinlichkeitstheorie verw. Geb., Vol. 67, 1984, p. 1-62.

[4] S. E. GRAVERSEN and J. VUOLLE-APIALA, 03B1-Self-Similar Markov Processes, Prob. Th.

Rel. Fields, Vol. 71, 1986, pp. 149-158.

[5] J. W. LAMPERTI, Semi-Stable Markov Processes I, Z. Wahrscheinlichkeitstheorie verw.

Geb., Vol. 22, 1972, pp. 205-225.

[6] J. VUOLLE-APIALA and S. E. GRAVERSEN, Duality Theory for Self-Similar Processes, Ann. Inst. Henri Poincaré, Probabilités et Statistiques, Vol. 22, (3), 1986, pp. 323-332.

(Manuscript received November 22th, 1988) (accepted June 15th, 1989.)

Références

Documents relatifs

Under symmetrical pattern, the deviation is in the domain of a large waiting time: a large time gap between successive rare events occurs more frequency then the exponential

I shall obviously not forget our Scientific Committee and the Sorbonne Law School and in particular its director and the administrators of the external relations

Nous n’oublions évidemment pas les membres de notre comité scientifique ainsi que l’École de droit de la Sorbonne et notamment son directeur et les administratrices du

Multi- self-similar Markov processes were introduced by Jacobsen and Yor [11] in the aim of extending the Lamperti transformation of positive self-similar Markov processes to R d

Key words and phrases: Positive self-similar Markov processes, Lamperti representation, conditioned stable L´evy processes, first exit time, first hitting time, exponential

Key words: Non-local diffusion; Time delay; Self-similar behaviour; Fractional Sobolev estimates; Central Limit Theorem..

The only sssi Gaussian process is the fractional Brownian motion; however there are many different sssi symmetric stable processes, which often arise in limit theorems for the

In [2], the authors present a discrete schema, called the random rewards schema, and show its convergence to a stable self-similar statinonary incre- ments process called